Barcelona, 8-9 September 2011 A Graph Visualization Tool for Terminology Discovery and Assessment Benoît Robichaud Observatoire de linguistique Sens-Texte (OLST) Université de Montréal [email protected] Abstract This paper presents a Graphical User Interface (GUI) mainly based on a graph visualization device and used for exploring and assessing lexical data found in the DiCoInfo, a specialized e-dictionary of computing and the Internet. Computer visualization devices have been used to present and browse data in many fields, but GUIs for electronic dictionaries have not evolved much. Very few take advantage of the fundamental nature of dictionaries: they are huge and ordered collections of lexical relationships (i.e. lexical networks). Graph visualization devices such as intertwined (directed) graphs present themselves as better tools to browse these relationships. They surely are well suited for assessing the consistency of encoded data. Keywords Lexical relations, e-dictionary, data visualization, graph model, assessment tool. 1 Introduction* Electronic support to dictionary content management has changed a great deal how data are encoded, managed and retrieved, but little work has been done on innovative ways to give end users ‘a richer experience’. For more than two decades, computer visualization devices have been set up to present and browse data from a multitude of sources and in many fields, but most current electronic dictionaries (e-dictionaries) merely continue to replicate the layout of their traditional printed counterparts to display their contents. Aside from image-based dictionaries that are notorious exceptions (for example: the Merriam-Webster’s Visual Dictionary Online, QAI’s The Visual Dictionary), many advantages of computer capabilities for data visualization have yet to be acquired and adapted in this field. This paper presents the goals, architecture and usability of a prototypical Graphical User Interface (GUI) primarily based on a graph visualization device and used to browse data and discover knowledge through a subset of selected relations that are found in the DiCoInfo (i.e. Dictionnaire fondamental de l’informatique et de l’Internet), an online specialized * We would like to thank M.-C. L’Homme and G. Bernier-Colborne from the OLST for very helpful suggestions and comments on an earlier draft of this paper. 243 Benoît Robichaud e-dictionary of computing and the Internet. This particular project is part of a larger effort to improve data and knowledge access for language professionals such as technical writers and translators (see L’Homme & Leroyer, 2009; L’Homme et al., 2010). Its birth is linked to the idea that it was possible to improve the visual and communicative value of dictionary contents using a graph visualization device. First, in displaying the links between the data that appear in field entries: for example, the lexical relationships that exist otherwise among synonyms, derivatives and related meanings of a particular term. Second, in displaying the links between entries that share particular data in some field entry: for example, the relationships among records that mention a particular term as a derivative or related meaning. Not only do these enhancements seem beneficial, they may be brought together in a single generalized representation that remains neutral with regard to the way the data is accessed. Figure 1 shows the kind of data visualization one can expect to obtain with this approach: Figure 1: Some of the lexical relations of the polysemous French term ‘exécuter’ The actual project was undertaken for two main reasons: 1. We assumed that relationships between terms (perhaps not all, but a large part of them) were likely to be better understood by end users if they were first shown graphically rather than simply listed in tables with textual explanations. In terminology, taxonomies and meronymies are usually presented in a graphical hierarchy, but other relationships could also lend themselves to a graphical presentation. 2. We also sought to offer a tool for terminologists updating the entries that would help them better assess the consistency of the descriptions. For instance, bidirectional relationships such as synonyms, antonyms, derivatives and related meanings could be more easily assessed using a graphical interface. The rest of the article is organized as follows. Section 2 presents a short overview of traditional GUIs to e-dictionaries and discusses specific drawbacks. It also gives a brief description of a few graph-based GUIs that are found online or downloadable from the Internet. Section 3 first briefly presents the DiCoInfo, and then provides technical details on the architecture and the features implemented so far in our graph-based GUI. Section 4 discusses directions for future work and some of the challenges they raise. Finally, a few concluding remarks are given in Section 5. 244 A Graph Visualization Tool for Terminology Discovery and Assessment 2 Old and new ways to explore dictionaries As previously mentioned, most GUIs to e-dictionaries merely continue to offer traditional outlooks on their contents. Few of them have only the mandatory nomenclature and a display mechanism to view chosen records. Many GUIs offer advanced general and ‘by-category’ search capabilities that produce (sometimes dynamically) shorter wordlists to help end users access specific contents more efficiently (see for example, Larousse, 2011; Le Petit Robert, 2011; OED, 2011). But wordlists are always presented in the very natural but immutable alphabetically-ordered fashion without showing the links between results. As Manning et al. (2001) mentioned, the basic reason seems to be that, contrary to encyclopedias and thesauri that organize their contents primarily on a conceptual basis, e-dictionaries always compile their contents solely as indexes. Another fundamental reason is simply that they organize and show search results only with respect to field entry organization. A last reason might be that despite the fact that they provide relationships between lexical units, very few encode these relationships formally (however, see Miller, 1993 and Steinlin et al., 2005). As Polguère (2009) puts it, the vast majority are simply text-based e-dictionaries, that is they only index field entry data and do not organize them otherwise. Nonetheless, during the last decade, innovative means for exploring e-dictionaries for end users have been proposed. Some of them rely predominantly on lexical networks and offer appealing and interactive graph visualization devices to navigate within their content (e.g., Jansz et al.’s Kirrkirr, 2008; The LexiCon Research Group’s EcoLexicon, 2009; Thinkmap’s Visual Thesaurus, 2011; logicalOctopus’s Visuwords, 2011; Vercruysse’s WordVis, 2011). However, without appropriate additional control options or display features (like drawing options that allow to select relationship types, see Section 3.2), these GUIs can quickly become confusing and users may have trouble untangling all the information presented. Figure 2 Figure 3 Lexical relations of the English word Lexical relations of the English word ‘save’ from Thinkmap’s Visual Thesaurus ‘computer’ from logicalOctopus’s Visuwords 3 The DiCoInfo and the DiCoInfo Visuel As mentioned in the first section, the DiCoInfo is an online e-dictionary that describes terms in the fields of computing and the Internet in French, English and Spanish. It was originally developed as a monolingual tool with the main function of helping end users solve specific knowledge problems associated with this specialized language. From year to year, new 245 Benoît Robichaud languages and functionalities have been added to assist them with tasks such as translation and text production in a second language (see L’Homme et al., 2009). 3.1 The DiCoInfo terminological database The records of the DiCoInfo are encoded in XML files that are stored in an eXist database management system (see Meier et al., 2011). Apart from the new graph-based GUI presented in Section 3.2 below, end users access and browse the dictionary contents via two main Web interfaces. The first one, called the static version, is a compilation of hyperlinked HTML pages that provides the list of all records in the conventional alphabetical fashion. The second one is a search version that mimics a search engine and finds the records containing strings (corresponding to parts of words or terms) in specific field entries such as the usual headword, variants and synonyms, but also in other fields that group different sorts (or families) of paradigmatic and syntagmatic lexical relationships. These last relationships are formally classified and encoded by means of the lexical functions used in the Explanatory Combinatorial Lexicology framework (see Mel’čuk et al., 1995 and Mel’čuk, 1996). Both GUIs are implemented using customary XSLT stylesheets that transform the original XML records and put them together in HTML format (see Clark, 1999). The next subsection describes the architecture of a new graph-based GUI designed for the DiCoInfo. Technical details are provided on the features that have been implemented so far. It is worth mentioning that subsets of lexical functions used in the DiCoInfo were specifically selected for this first version. These encode paradigmatic relationships, namely hypernyms, synonyms, antonyms, derivatives and related meanings. Hyponymic and meronymic relationships are not yet incorporated since the data themselves need to be revised and